[ovs-dev] BUG: Weird failure of OVS classifier

[Ben Pfaff]

On Wed, Mar 25, 2020 at 07:35:22PM -0700, Jean Tourrilhes wrote:
> 	Hi all,
> 
> 	I have stumbled on a weird failure of the OVS classifier. This
> is a bit beyond my abilities, so I would like other people to have
> a look at it.
> 	This is the error in the log :
> -------------------------------------------
> 2020-03-26T02:05:22.693Z|01237|dpif(handler86)|DBG|system at ovs-system: failed to put[create] (File exists) ufid:a4bba536-21ad-4c5d-9dd9-ab8f72263ce1 recirc_id(0),dp_hash(0/0),skb_priority(0/0),in_port(1),skb_mark(0/0),ct_state(0x2a/0),ct_zone(0/0),ct_mark(0/0),ct_label(0/0),ct_tuple4(src=10.68.58.40/0.0.0.0,dst=10.68.58.41/0.0.0.0,proto=17/0,tp_src=55732/0,tp_dst=7364/0),eth(src=22:c3:19:dc:23:47/00:00:00:00:00:00,dst=fa:c1:63:72:87:45/00:00:00:00:00:00),eth_type(0x0800),ipv4(src=10.68.58.41/0.0.0.0,dst=10.68.58.40/255.255.255.0,proto=17,tos=0/0x3,ttl=64/0,frag=no),udp(src=7364/0,dst=55732/0x8000), actions:set(tunnel(tun_id=0x0,dst=10.68.68.40,ttl=64,tp_dst=4789,flags(df|key))),2
> -------------------------------------------
> 	From my understanding, this should not happen, so one of the
> classifier is wrong, the main question is if it is the kernel
> classifier or the vswitchd classifier.

Usually this kind of thing means that userspace and the kernel don't
have the same idea of the definition of a flow.  Tracking it down is
often tricky.

This kind of miserable problem is one reason why I wish that OVS could
switch to an entirely packet-based interface.  The user/kernel interface
is almost unmaintainable.